2008/5/26 Henrik Holst <[EMAIL PROTECTED]>:
>
> We have performed some quick benchmarks against kvm-69 on an Intel Q6600
> with 8GB RAM and equipped with 4 SATA drives (Maxtor DiamondMax 10
> 6V300F0).
>
> During these tests we noticed a problem with the virtio block device in
> where it completely locks the drive after a while when running on a smp
> guest. The other disks are ok and the guest is working as it should but
> all access to /dev/vda1 never times out.
>
> Also with if=scsi the host would output (during the bonnie++ tests):
> lsi_scsi: error: PRDERED queue not implemented
> lsi_scsi: error: Unimplemented message 0x0c
>
> Host: 2.6.24-1-amd64 debian testing
> Guest: 2.6.24-16-server ubuntu 8.04
>
> hdparm -tT /dev/sdb on host:
> Timing cached reads: 3845.63 MB/sec
> Timing buffered disk reads: 67.98 MB/sec
>
> Qemu invocation (for virtio):
> qemu-system-x86_64
> -m 3072 -smp 4 \
> -drive file=ubuntu.img,if=ide,boot=on,media=disk \
> -drive file=/dev/sdb,if=virtio,boot=off,cache=off,media=disk
>
> Bonnie++ invocation:
> bonnie++ -r 512 -s 2048 -u root -d /mnt/sdb1/ :
>
> Results:
> ---------------------------------------------------------------------------------
> Version 1.03b ------Sequential Output------ --Sequential Input-
> --Random-
> -Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
> Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec
> %CP
> host 2G 60814 81 83149 15 64419 8 85187 99 2394159 100 +++++
> +++
> guest-ide 2G 65048 89 72621 13 64792 9 84224 99 2311183 99 +++++
> +++
> guest-scsi 2G 67387 90 73757 13 63783 9 84645 100 2297687 99 +++++
> +++
> guest-virtio 2G 41564 57 73274 28 62725 27 71656 85 2441544 98 +++++
> +++ (This is with -smp 1)
>
> There seams to be no major performance difference between ide and scsi while
> virtio is not to be recommended at this moment.
> All in all the IO looks very close to native, at least with this limited
> tests.
>
> /Henrik Holst
>
>
I've just re-run this on kvm-71 and virtio seams to be corrected on
smp systems now. We could find no lockups using virtio and 4 smp
cores.
Version 1.03b ------Sequential Output------ --Sequential Input- --Random-
-Per Chr- --Block-- -Rewrite- -Per Chr- --Block-- --Seeks--
Machine Size K/sec %CP K/sec %CP K/sec %CP K/sec %CP K/sec %CP /sec %CP
host 2G 60814 81 83149 15 64419 8 85187 99 2394159
100 +++++ +++
guest-ide 2G 65048 89 72621 13 64792 9 84224 99 2311183 99 +++++ +++
guest-virtio 2G 44011 68 92460 32 34272 9 38224 47 2277747 99 +++++ +++
Virtio block devices seams thus to be slower on putc and reads overall
while faster on block writes. For some reason we couldn't manage to
boot a scsi guest on the smp host, I'll get back with more info on
this as soon as I have gathered some.
/Henrik Holst
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
